Diploma and Master Theses (authored and supervised):
"Dependable Event Processing over High Volume Data Streams";
Supervisor: S. Dustdar, W. Hummer;
Institut für Informationssysteme, Distributed Systems Group,
final examination: 2014-04-24.
The number of event processing systems is increasing more and more. In contrast to traditional systems, those event processing systems do not handle persistent data, which is mostly stored in databases, but instead they have to deal with events, which are received continuously over various communication channels and should be processed more or less immediately. Complex computations (e.g., mathematical calculations, or pattern detection) in terms of queries have to be performed for these events. The frequency of the arriving events is not necessarily steady. In fact, the system has do deal with ups and downs, which can influence the data volume heavilyload
fluctuations, which can heavily influence the requirements concerning processing power. Especially in times of high data volume the processing systems have to deal with a huge number of events and should be able to manage these phases.
Different approaches of dealing with high volume data streams have been studied, but their applicability and efficiency may vary depending on the application scenario. General approaches or guidelines on how to treat such overload are not available.
This thesis covers strategies to handle phases of high data volume on event streams for a single event processing node. Currently, there are three established strategies for coping with loads that are too high, which have been used to treat overload situations caused by high data volume: load shedding, deferred execution and forwarding. These strategies are discussed and their applicability for different types of queries is evaluated. To that end, a taxonomy of queries
in event processing systems is elaborated. The taxonomy covers different dimensions like the type of processing operation and the scope of the query.
Based on the features of the strategies and the different query types, the applicability of the strategies is analyzed in theory and an evaluation is performed to support the analysis. The strategies are implemented in a generic way and are integrated into the WS-Aggregation framework for the evaluation. This framework for distributed and event-based aggregation of web services data has been developed by the Distributed Systems Group at the Vienna University of
Technology. Furthermore, the results of the evaluation are used to determine the strength of the influence of the different applicability criteria and to formulate problem statements for further
Created from the Publication Database of the Vienna University of Technology.